Deep learning-assisted ultrathin broadband vanadium dioxide-based polarization-insensitive terahertz metamaterial absorber.

In this article, we presented a broadband absorber structure design consist of double arrays of VO2 based complementary plus placed on top of a polyimide dielectric spacer. The absorber exhibits a compact design and ultrathin thickness of 0.328 λ and 0.062 λ, respectively, and achieves absorptivity above 90% over the 2.66–6.48 THz range, corresponding to a relative bandwidth of 83.6%. The physical mechanism of broadband absorption is explained by parametric analysis with electric, magnetic field, and surface current distributions. Furthermore, the proposed absorber has wide incident angle absorption and polarization insensitivity due to its symmetrical structure. To optimize the absorption performance, Deep learning techniques such as Multi-Layer Perceptron (MLP), Residual MLP, and Deep & Cross Network (DCN) models are introduced. Prediction performance is evaluated using 5- and 10-fold cross-validation, with the Residual MLP achieving the highest R2 and lowest error metrics. The proposed broadband absorber with deep learning has potential applications in electromagnetic stealth, cloaking, modulators, detection, sensing, and energy harvesting.